Machine and Deep Learning
Anthony C. Chang · Intelligence-Based Medicine · 2020
Machine learning along with data mining comprise sub-disciplines under data science. There are several schools of machine learning, including symbolists, connectionists, revolutionaries, Bayesians, and analogizers. Machine learning with its sometimes tedious workflow differs significantly from conventional programming. Classical machine learning consists of supervised (classification and regression) and unsupervised (clustering and generalization) learning but also semi-supervised and ensemble learning. Deep learning consists of a range of methods including convolutional neural networks, recurrent neural networks, generative adversarial networks, and their derivatives. Deep reinforcement learning (such as deep Q network, or DQN) is becoming a valuable deep learning tool in biomedicine. Evaluation of these models includes methods such as receiver operating characteristic, precision-recall curve, and the F-1 measure in the confusion matrix. Finally, issues such as explainability, bias and variance, fitting, curse of dimensionality, and correlation vs causation are discussed.